Commit e7270e47 authored by Linus Torvalds's avatar Linus Torvalds
Browse files

Merge tag 's390-5.12-2' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux 

Pull more s390 updates from Vasily Gorbik:

 - Fix physical vs virtual confusion in some basic mm macros and
   routines. Caused by __pa == __va on s390 currently.

 - Get rid of on-stack cpu masks.

 - Add support for complete CPU counter set extraction.

 - Add arch_irq_work_raise implementation.

 - virtio-ccw revision and opcode fixes.

* tag 's390-5.12-2' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/cpumf: Add support for complete counter set extraction
  virtio/s390: implement virtio-ccw revision 2 correctly
  s390/smp: implement arch_irq_work_raise()
  s390/topology: move cpumasks away from stack
  s390/smp: smp_emergency_stop() - move cpumask away from stack
  s390/smp: __smp_rescan_cpus() - move cpumask away from stack
  s390/smp: consolidate locking for smp_rescan()
  s390/mm: fix phys vs virt confusion in vmem_*() functions family
  s390/mm: fix phys vs virt confusion in pgtable allocation routines
  s390/mm: fix invalid __pa() usage in pfn_pXd() macros
  s390/mm: make pXd_deref() macros return a pointer
  s390/opcodes: rename selhhhr to selfhr
parents c19798af cf6acb8b

arch/s390/include/asm/irq_work.h

0 → 100644
+12 −0
Original line number Diff line number Diff line 
/* SPDX-License-Identifier: GPL-2.0 */

#ifndef _ASM_S390_IRQ_WORK_H

#define _ASM_S390_IRQ_WORK_H



static inline bool arch_irq_work_has_interrupt(void)

{

	return true;

}



void arch_irq_work_raise(void);



#endif /* _ASM_S390_IRQ_WORK_H */

arch/s390/include/asm/pgalloc.h

+1 −1
Original line number Diff line number Diff line
@@ -135,7 +135,7 @@ static inline void pmd_populate(struct mm_struct *mm, 
#define pmd_populate_kernel(mm, pmd, pte) pmd_populate(mm, pmd, pte)



#define pmd_pgtable(pmd) \

	(pgtable_t)(pmd_val(pmd) & -sizeof(pte_t)*PTRS_PER_PTE)

	((pgtable_t)__va(pmd_val(pmd) & -sizeof(pte_t)*PTRS_PER_PTE))



/*

 * page table entry allocation/free routines.

arch/s390/include/asm/pgtable.h

+8 −8
Original line number Diff line number Diff line
@@ -1219,8 +1219,8 @@ static inline pte_t mk_pte(struct page *page, pgprot_t pgprot) 
#define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))

#define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))



#define p4d_deref(pud) (p4d_val(pud) & _REGION_ENTRY_ORIGIN)

#define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)

#define p4d_deref(pud) ((unsigned long)__va(p4d_val(pud) & _REGION_ENTRY_ORIGIN))

#define pgd_deref(pgd) ((unsigned long)__va(pgd_val(pgd) & _REGION_ENTRY_ORIGIN))



static inline unsigned long pmd_deref(pmd_t pmd)

{
@@ -1229,12 +1229,12 @@ static inline unsigned long pmd_deref(pmd_t pmd) 
	origin_mask = _SEGMENT_ENTRY_ORIGIN;

	if (pmd_large(pmd))

		origin_mask = _SEGMENT_ENTRY_ORIGIN_LARGE;

	return pmd_val(pmd) & origin_mask;

	return (unsigned long)__va(pmd_val(pmd) & origin_mask);

}



static inline unsigned long pmd_pfn(pmd_t pmd)

{

	return pmd_deref(pmd) >> PAGE_SHIFT;

	return __pa(pmd_deref(pmd)) >> PAGE_SHIFT;

}



static inline unsigned long pud_deref(pud_t pud)
@@ -1244,12 +1244,12 @@ static inline unsigned long pud_deref(pud_t pud) 
	origin_mask = _REGION_ENTRY_ORIGIN;

	if (pud_large(pud))

		origin_mask = _REGION3_ENTRY_ORIGIN_LARGE;

	return pud_val(pud) & origin_mask;

	return (unsigned long)__va(pud_val(pud) & origin_mask);

}



static inline unsigned long pud_pfn(pud_t pud)

{

	return pud_deref(pud) >> PAGE_SHIFT;

	return __pa(pud_deref(pud)) >> PAGE_SHIFT;

}



/*
@@ -1329,7 +1329,7 @@ static inline bool gup_fast_permitted(unsigned long start, unsigned long end) 
}

#define gup_fast_permitted gup_fast_permitted



#define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))

#define pfn_pte(pfn, pgprot)	mk_pte_phys(((pfn) << PAGE_SHIFT), (pgprot))

#define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)

#define pte_page(x) pfn_to_page(pte_pfn(x))
@@ -1636,7 +1636,7 @@ static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma, 
}

#define pmdp_collapse_flush pmdp_collapse_flush



#define pfn_pmd(pfn, pgprot)	mk_pmd_phys(__pa((pfn) << PAGE_SHIFT), (pgprot))

#define pfn_pmd(pfn, pgprot)	mk_pmd_phys(((pfn) << PAGE_SHIFT), (pgprot))

#define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))



static inline int pmd_trans_huge(pmd_t pmd)

arch/s390/include/uapi/asm/perf_cpum_cf_diag.h

0 → 100644
+51 −0
Original line number Diff line number Diff line 
/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */

/*

 * Copyright IBM Corp. 2021

 * Interface implementation for communication with the CPU Measurement

 * counter facility device driver.

 *

 * Author(s): Thomas Richter <tmricht@linux.ibm.com>

 *

 * Define for ioctl() commands to communicate with the CPU Measurement

 * counter facility device driver.

 */



#ifndef _PERF_CPUM_CF_DIAG_H

#define _PERF_CPUM_CF_DIAG_H



#include <linux/ioctl.h>

#include <linux/types.h>



#define S390_HWCTR_DEVICE		"hwctr"

#define S390_HWCTR_START_VERSION	1



struct s390_ctrset_start {		/* Set CPUs to operate on */

	__u64 version;			/* Version of interface */

	__u64 data_bytes;		/* # of bytes required */

	__u64 cpumask_len;		/* Length of CPU mask in bytes */

	__u64 *cpumask;			/* Pointer to CPU mask */

	__u64 counter_sets;		/* Bit mask of counter sets to get */

};



struct s390_ctrset_setdata {		/* Counter set data */

	__u32 set;			/* Counter set number */

	__u32 no_cnts;			/* # of counters stored in cv[] */

	__u64 cv[0];			/* Counter values (variable length) */

};



struct s390_ctrset_cpudata {		/* Counter set data per CPU */

	__u32 cpu_nr;			/* CPU number */

	__u32 no_sets;			/* # of counters sets in data[] */

	struct s390_ctrset_setdata data[0];

};



struct s390_ctrset_read {		/* Structure to get all ctr sets */

	__u64 no_cpus;			/* Total # of CPUs data taken from */

	struct s390_ctrset_cpudata data[0];

};



#define S390_HWCTR_MAGIC	'C'	/* Random magic # for ioctls */

#define	S390_HWCTR_START	_IOWR(S390_HWCTR_MAGIC, 1, struct s390_ctrset_start)

#define	S390_HWCTR_STOP		_IO(S390_HWCTR_MAGIC, 2)

#define	S390_HWCTR_READ		_IOWR(S390_HWCTR_MAGIC, 3, struct s390_ctrset_read)

#endif

arch/s390/kernel/perf_cpum_cf_diag.c

+525 −23
Original line number Diff line number Diff line
@@ -2,7 +2,7 @@ 
/*

 * Performance event support for s390x - CPU-measurement Counter Sets

 *

 *  Copyright IBM Corp. 2019

 *  Copyright IBM Corp. 2019, 2021

 *  Author(s): Hendrik Brueckner <brueckner@linux.ibm.com>

 *	       Thomas Richer <tmricht@linux.ibm.com>

 */
@@ -17,6 +17,8 @@ 
#include <linux/export.h>

#include <linux/slab.h>

#include <linux/processor.h>

#include <linux/miscdevice.h>

#include <linux/mutex.h>



#include <asm/ctl_reg.h>

#include <asm/irq.h>
@@ -24,8 +26,12 @@ 
#include <asm/timex.h>

#include <asm/debug.h>



#define	CF_DIAG_CTRSET_DEF		0xfeef	/* Counter set header mark */

#include <asm/perf_cpum_cf_diag.h>



#define	CF_DIAG_CTRSET_DEF		0xfeef	/* Counter set header mark */

#define CF_DIAG_MIN_INTERVAL		60	/* Minimum counter set read */

						/* interval in seconds */

static unsigned long cf_diag_interval = CF_DIAG_MIN_INTERVAL;

static unsigned int cf_diag_cpu_speed;

static debug_info_t *cf_diag_dbg;
@@ -33,6 +39,7 @@ struct cf_diag_csd { /* Counter set data per CPU */ 
	size_t used;			/* Bytes used in data/start */

	unsigned char start[PAGE_SIZE];	/* Counter set at event start */

	unsigned char data[PAGE_SIZE];	/* Counter set at event delete */

	unsigned int sets;		/* # Counter set saved in data */

};

static DEFINE_PER_CPU(struct cf_diag_csd, cf_diag_csd);
@@ -178,18 +185,35 @@ static void cf_diag_disable(struct pmu *pmu) 


/* Number of perf events counting hardware events */

static atomic_t cf_diag_events = ATOMIC_INIT(0);

/* Used to avoid races in calling reserve/release_cpumf_hardware */

static DEFINE_MUTEX(cf_diag_reserve_mutex);



/* Release the PMU if event is the last perf event */

static void cf_diag_perf_event_destroy(struct perf_event *event)

{

	debug_sprintf_event(cf_diag_dbg, 5,

			    "%s event %p cpu %d cf_diag_events %d\n",

			    __func__, event, event->cpu,

			    __func__, event, smp_processor_id(),

			    atomic_read(&cf_diag_events));

	if (atomic_dec_return(&cf_diag_events) == 0)

		__kernel_cpumcf_end();

}



static int get_authctrsets(void)

{

	struct cpu_cf_events *cpuhw;

	unsigned long auth = 0;

	enum cpumf_ctr_set i;



	cpuhw = &get_cpu_var(cpu_cf_events);

	for (i = CPUMF_CTR_SET_BASIC; i < CPUMF_CTR_SET_MAX; ++i) {

		if (cpuhw->info.auth_ctl & cpumf_ctr_ctl[i])

			auth |= cpumf_ctr_ctl[i];

	}

	put_cpu_var(cpu_cf_events);

	return auth;

}



/* Setup the event. Test for authorized counter sets and only include counter

 * sets which are authorized at the time of the setup. Including unauthorized

 * counter sets result in specification exception (and panic).
@@ -197,15 +221,12 @@ static void cf_diag_perf_event_destroy(struct perf_event *event) 
static int __hw_perf_event_init(struct perf_event *event)

{

	struct perf_event_attr *attr = &event->attr;

	struct cpu_cf_events *cpuhw;

	enum cpumf_ctr_set i;

	int err = 0;



	debug_sprintf_event(cf_diag_dbg, 5, "%s event %p cpu %d\n", __func__,

			    event, event->cpu);



	event->hw.config = attr->config;

	event->hw.config_base = 0;



	/* Add all authorized counter sets to config_base. The

	 * the hardware init function is either called per-cpu or just once
@@ -215,11 +236,7 @@ static int __hw_perf_event_init(struct perf_event *event) 
	 * Checking the authorization on any CPU is fine as the hardware

	 * applies the same authorization settings to all CPUs.

	 */

	cpuhw = &get_cpu_var(cpu_cf_events);

	for (i = CPUMF_CTR_SET_BASIC; i < CPUMF_CTR_SET_MAX; ++i)

		if (cpuhw->info.auth_ctl & cpumf_ctr_ctl[i])

			event->hw.config_base |= cpumf_ctr_ctl[i];

	put_cpu_var(cpu_cf_events);

	event->hw.config_base = get_authctrsets();



	/* No authorized counter sets, nothing to count/sample */

	if (!event->hw.config_base) {
@@ -237,6 +254,25 @@ static int __hw_perf_event_init(struct perf_event *event) 
	return err;

}



/* Return 0 if the CPU-measurement counter facility is currently free

 * and an error otherwise.

 */

static int cf_diag_perf_event_inuse(void)

{

	int err = 0;



	if (!atomic_inc_not_zero(&cf_diag_events)) {

		mutex_lock(&cf_diag_reserve_mutex);

		if (atomic_read(&cf_diag_events) == 0 &&

		    __kernel_cpumcf_begin())

			err = -EBUSY;

		else

			err = atomic_inc_return(&cf_diag_events);

		mutex_unlock(&cf_diag_reserve_mutex);

	}

	return err;

}



static int cf_diag_event_init(struct perf_event *event)

{

	struct perf_event_attr *attr = &event->attr;
@@ -264,13 +300,9 @@ static int cf_diag_event_init(struct perf_event *event) 
	}



	/* Initialize for using the CPU-measurement counter facility */

	if (atomic_inc_return(&cf_diag_events) == 1) {

		if (__kernel_cpumcf_begin()) {

			atomic_dec(&cf_diag_events);

			err = -EBUSY;

	err = cf_diag_perf_event_inuse();

	if (err < 0)

		goto out;

		}

	}

	event->destroy = cf_diag_perf_event_destroy;



	err = __hw_perf_event_init(event);
@@ -599,6 +631,8 @@ static void cf_diag_del(struct perf_event *event, int flags) 
	cpuhw->flags &= ~PMU_F_IN_USE;

}



/* Default counter set events and format attribute groups */



CPUMF_EVENT_ATTR(CF_DIAG, CF_DIAG, PERF_EVENT_CPUM_CF_DIAG);



static struct attribute *cf_diag_events_attr[] = {
@@ -663,6 +697,452 @@ static void cf_diag_get_cpu_speed(void) 
	}

}



/* Code to create device and file I/O operations */

static atomic_t ctrset_opencnt = ATOMIC_INIT(0);	/* Excl. access */



static int cf_diag_open(struct inode *inode, struct file *file)

{

	int err = 0;



	if (!capable(CAP_SYS_ADMIN))

		return -EPERM;

	if (atomic_xchg(&ctrset_opencnt, 1))

		return -EBUSY;



	/* Avoid concurrent access with perf_event_open() system call */

	mutex_lock(&cf_diag_reserve_mutex);

	if (atomic_read(&cf_diag_events) || __kernel_cpumcf_begin())

		err = -EBUSY;

	mutex_unlock(&cf_diag_reserve_mutex);

	if (err) {

		atomic_set(&ctrset_opencnt, 0);

		return err;

	}

	file->private_data = NULL;

	debug_sprintf_event(cf_diag_dbg, 2, "%s\n", __func__);

	/* nonseekable_open() never fails */

	return nonseekable_open(inode, file);

}



/* Variables for ioctl() interface support */

static DEFINE_MUTEX(cf_diag_ctrset_mutex);

static struct cf_diag_ctrset {

	unsigned long ctrset;		/* Bit mask of counter set to read */

	cpumask_t mask;			/* CPU mask to read from */

	time64_t lastread;		/* Epoch counter set last read */

} cf_diag_ctrset;



static void cf_diag_ctrset_clear(void)

{

	cpumask_clear(&cf_diag_ctrset.mask);

	cf_diag_ctrset.ctrset = 0;

}



static void cf_diag_release_cpu(void *p)

{

	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);



	debug_sprintf_event(cf_diag_dbg, 3, "%s cpu %d\n", __func__,

			    smp_processor_id());

	lcctl(0);		/* Reset counter sets */

	cpuhw->state = 0;	/* Save state in CPU hardware state */

}



/* Release function is also called when application gets terminated without

 * doing a proper ioctl(..., S390_HWCTR_STOP, ...) command.

 * Since only one application is allowed to open the device, simple stop all

 * CPU counter sets.

 */

static int cf_diag_release(struct inode *inode, struct file *file)

{

	on_each_cpu(cf_diag_release_cpu, NULL, 1);

	cf_diag_ctrset_clear();

	atomic_set(&ctrset_opencnt, 0);

	__kernel_cpumcf_end();

	debug_sprintf_event(cf_diag_dbg, 2, "%s\n", __func__);

	return 0;

}



struct cf_diag_call_on_cpu_parm {	/* Parm struct for smp_call_on_cpu */

	unsigned int sets;		/* Counter set bit mask */

	atomic_t cpus_ack;		/* # CPUs successfully executed func */

};



static int cf_diag_all_copy(unsigned long arg, cpumask_t *mask)

{

	struct s390_ctrset_read __user *ctrset_read;

	unsigned int cpu, cpus, rc;

	void __user *uptr;



	ctrset_read = (struct s390_ctrset_read __user *)arg;

	uptr = ctrset_read->data;

	for_each_cpu(cpu, mask) {

		struct cf_diag_csd *csd = per_cpu_ptr(&cf_diag_csd, cpu);

		struct s390_ctrset_cpudata __user *ctrset_cpudata;



		ctrset_cpudata = uptr;

		debug_sprintf_event(cf_diag_dbg, 5, "%s cpu %d used %zd\n",

				    __func__, cpu, csd->used);

		rc  = put_user(cpu, &ctrset_cpudata->cpu_nr);

		rc |= put_user(csd->sets, &ctrset_cpudata->no_sets);

		rc |= copy_to_user(ctrset_cpudata->data, csd->data, csd->used);

		if (rc)

			return -EFAULT;

		uptr += sizeof(struct s390_ctrset_cpudata) + csd->used;

		cond_resched();

	}

	cpus = cpumask_weight(mask);

	if (put_user(cpus, &ctrset_read->no_cpus))

		return -EFAULT;

	debug_sprintf_event(cf_diag_dbg, 5, "%s copied %ld\n",

			    __func__, uptr - (void __user *)ctrset_read->data);

	return 0;

}



static size_t cf_diag_cpuset_read(struct s390_ctrset_setdata *p, int ctrset,

				  int ctrset_size, size_t room)

{

	size_t need = 0;

	int rc = -1;



	need = sizeof(*p) + sizeof(u64) * ctrset_size;

	debug_sprintf_event(cf_diag_dbg, 5,

			    "%s room %zd need %zd set %#x set_size %d\n",

			    __func__, room, need, ctrset, ctrset_size);

	if (need <= room) {

		p->set = cpumf_ctr_ctl[ctrset];

		p->no_cnts = ctrset_size;

		rc = ctr_stcctm(ctrset, ctrset_size, (u64 *)p->cv);

		if (rc == 3)		/* Nothing stored */

			need = 0;

	}

	debug_sprintf_event(cf_diag_dbg, 5, "%s need %zd rc %d\n", __func__,

			    need, rc);

	return need;

}



/* Read all counter sets. Since the perf_event_open() system call with

 * event cpum_cf_diag/.../ is blocked when this interface is active, reuse

 * the perf_event_open() data buffer to store the counter sets.

 */

static void cf_diag_cpu_read(void *parm)

{

	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);

	struct cf_diag_csd *csd = this_cpu_ptr(&cf_diag_csd);

	struct cf_diag_call_on_cpu_parm *p = parm;

	int set, set_size;

	size_t space;



	debug_sprintf_event(cf_diag_dbg, 5,

			    "%s new %#x flags %#x state %#llx\n",

			    __func__, p->sets, cpuhw->flags,

			    cpuhw->state);

	/* No data saved yet */

	csd->used = 0;

	csd->sets = 0;

	memset(csd->data, 0, sizeof(csd->data));



	/* Scan the counter sets */

	for (set = CPUMF_CTR_SET_BASIC; set < CPUMF_CTR_SET_MAX; ++set) {

		struct s390_ctrset_setdata *sp = (void *)csd->data + csd->used;



		if (!(p->sets & cpumf_ctr_ctl[set]))

			continue;	/* Counter set not in list */

		set_size = cf_diag_ctrset_size(set, &cpuhw->info);

		space = sizeof(csd->data) - csd->used;

		space = cf_diag_cpuset_read(sp, set, set_size, space);

		if (space) {

			csd->used += space;

			csd->sets += 1;

		}

		debug_sprintf_event(cf_diag_dbg, 5, "%s sp %px space %zd\n",

				    __func__, sp, space);

	}

	debug_sprintf_event(cf_diag_dbg, 5, "%s sets %d used %zd\n", __func__,

			    csd->sets, csd->used);

}



static int cf_diag_all_read(unsigned long arg)

{

	struct cf_diag_call_on_cpu_parm p;

	cpumask_var_t mask;

	time64_t now;

	int rc = 0;



	debug_sprintf_event(cf_diag_dbg, 5, "%s\n", __func__);

	if (!alloc_cpumask_var(&mask, GFP_KERNEL))

		return -ENOMEM;

	now = ktime_get_seconds();

	if (cf_diag_ctrset.lastread + cf_diag_interval > now) {

		debug_sprintf_event(cf_diag_dbg, 5, "%s now %lld "

				    " lastread %lld\n", __func__, now,

				    cf_diag_ctrset.lastread);

		rc = -EAGAIN;

		goto out;

	} else {

		cf_diag_ctrset.lastread = now;

	}

	p.sets = cf_diag_ctrset.ctrset;

	cpumask_and(mask, &cf_diag_ctrset.mask, cpu_online_mask);

	on_each_cpu_mask(mask, cf_diag_cpu_read, &p, 1);

	rc = cf_diag_all_copy(arg, mask);

out:

	free_cpumask_var(mask);

	debug_sprintf_event(cf_diag_dbg, 5, "%s rc %d\n", __func__, rc);

	return rc;

}



/* Stop all counter sets via ioctl interface */

static void cf_diag_ioctl_off(void *parm)

{

	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);

	struct cf_diag_call_on_cpu_parm *p = parm;

	int rc;



	debug_sprintf_event(cf_diag_dbg, 5,

			    "%s new %#x flags %#x state %#llx\n",

			    __func__, p->sets, cpuhw->flags,

			    cpuhw->state);



	ctr_set_multiple_disable(&cpuhw->state, p->sets);

	ctr_set_multiple_stop(&cpuhw->state, p->sets);

	rc = lcctl(cpuhw->state);		/* Stop counter sets */

	if (!cpuhw->state)

		cpuhw->flags &= ~PMU_F_IN_USE;

	debug_sprintf_event(cf_diag_dbg, 5,

			    "%s rc %d flags %#x state %#llx\n", __func__,

			     rc, cpuhw->flags, cpuhw->state);

}



/* Start counter sets on particular CPU */

static void cf_diag_ioctl_on(void *parm)

{

	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);

	struct cf_diag_call_on_cpu_parm *p = parm;

	int rc;



	debug_sprintf_event(cf_diag_dbg, 5,

			    "%s new %#x flags %#x state %#llx\n",

			    __func__, p->sets, cpuhw->flags,

			    cpuhw->state);



	if (!(cpuhw->flags & PMU_F_IN_USE))

		cpuhw->state = 0;

	cpuhw->flags |= PMU_F_IN_USE;

	rc = lcctl(cpuhw->state);		/* Reset unused counter sets */

	ctr_set_multiple_enable(&cpuhw->state, p->sets);

	ctr_set_multiple_start(&cpuhw->state, p->sets);

	rc |= lcctl(cpuhw->state);		/* Start counter sets */

	if (!rc)

		atomic_inc(&p->cpus_ack);

	debug_sprintf_event(cf_diag_dbg, 5, "%s rc %d state %#llx\n",

			    __func__, rc, cpuhw->state);

}



static int cf_diag_all_stop(void)

{

	struct cf_diag_call_on_cpu_parm p = {

		.sets = cf_diag_ctrset.ctrset,

	};

	cpumask_var_t mask;



	if (!alloc_cpumask_var(&mask, GFP_KERNEL))

		return -ENOMEM;

	cpumask_and(mask, &cf_diag_ctrset.mask, cpu_online_mask);

	on_each_cpu_mask(mask, cf_diag_ioctl_off, &p, 1);

	free_cpumask_var(mask);

	return 0;

}



static int cf_diag_all_start(void)

{

	struct cf_diag_call_on_cpu_parm p = {

		.sets = cf_diag_ctrset.ctrset,

		.cpus_ack = ATOMIC_INIT(0),

	};

	cpumask_var_t mask;

	int rc = 0;



	if (!alloc_cpumask_var(&mask, GFP_KERNEL))

		return -ENOMEM;

	cpumask_and(mask, &cf_diag_ctrset.mask, cpu_online_mask);

	on_each_cpu_mask(mask, cf_diag_ioctl_on, &p, 1);

	if (atomic_read(&p.cpus_ack) != cpumask_weight(mask)) {

		on_each_cpu_mask(mask, cf_diag_ioctl_off, &p, 1);

		rc = -EIO;

	}

	free_cpumask_var(mask);

	return rc;

}



/* Return the maximum required space for all possible CPUs in case one

 * CPU will be onlined during the START, READ, STOP cycles.

 * To find out the size of the counter sets, any one CPU will do. They

 * all have the same counter sets.

 */

static size_t cf_diag_needspace(unsigned int sets)

{

	struct cpu_cf_events *cpuhw = this_cpu_ptr(&cpu_cf_events);

	size_t bytes = 0;

	int i;



	for (i = CPUMF_CTR_SET_BASIC; i < CPUMF_CTR_SET_MAX; ++i) {

		if (!(sets & cpumf_ctr_ctl[i]))

			continue;

		bytes += cf_diag_ctrset_size(i, &cpuhw->info) * sizeof(u64) +

			 sizeof(((struct s390_ctrset_setdata *)0)->set) +

			 sizeof(((struct s390_ctrset_setdata *)0)->no_cnts);

	}

	bytes = sizeof(((struct s390_ctrset_read *)0)->no_cpus) + nr_cpu_ids *

		(bytes + sizeof(((struct s390_ctrset_cpudata *)0)->cpu_nr) +

		     sizeof(((struct s390_ctrset_cpudata *)0)->no_sets));

	debug_sprintf_event(cf_diag_dbg, 5, "%s bytes %ld\n", __func__,

			    bytes);

	return bytes;

}



static long cf_diag_ioctl_read(unsigned long arg)

{

	struct s390_ctrset_read read;

	int ret = 0;



	debug_sprintf_event(cf_diag_dbg, 5, "%s\n", __func__);

	if (copy_from_user(&read, (char __user *)arg, sizeof(read)))

		return -EFAULT;

	ret = cf_diag_all_read(arg);

	debug_sprintf_event(cf_diag_dbg, 5, "%s ret %d\n", __func__, ret);

	return ret;

}



static long cf_diag_ioctl_stop(void)

{

	int ret;



	debug_sprintf_event(cf_diag_dbg, 5, "%s\n", __func__);

	ret = cf_diag_all_stop();

	cf_diag_ctrset_clear();

	debug_sprintf_event(cf_diag_dbg, 5, "%s ret %d\n", __func__, ret);

	return ret;

}



static long cf_diag_ioctl_start(unsigned long arg)

{

	struct s390_ctrset_start __user *ustart;

	struct s390_ctrset_start start;

	void __user *umask;

	unsigned int len;

	int ret = 0;

	size_t need;



	if (cf_diag_ctrset.ctrset)

		return -EBUSY;

	ustart = (struct s390_ctrset_start __user *)arg;

	if (copy_from_user(&start, ustart, sizeof(start)))

		return -EFAULT;

	if (start.version != S390_HWCTR_START_VERSION)

		return -EINVAL;

	if (start.counter_sets & ~(cpumf_ctr_ctl[CPUMF_CTR_SET_BASIC] |

				   cpumf_ctr_ctl[CPUMF_CTR_SET_USER] |

				   cpumf_ctr_ctl[CPUMF_CTR_SET_CRYPTO] |

				   cpumf_ctr_ctl[CPUMF_CTR_SET_EXT] |

				   cpumf_ctr_ctl[CPUMF_CTR_SET_MT_DIAG]))

		return -EINVAL;		/* Invalid counter set */

	if (!start.counter_sets)

		return -EINVAL;		/* No counter set at all? */

	cpumask_clear(&cf_diag_ctrset.mask);

	len = min_t(u64, start.cpumask_len, cpumask_size());

	umask = (void __user *)start.cpumask;

	if (copy_from_user(&cf_diag_ctrset.mask, umask, len))

		return -EFAULT;

	if (cpumask_empty(&cf_diag_ctrset.mask))

		return -EINVAL;

	need = cf_diag_needspace(start.counter_sets);

	if (put_user(need, &ustart->data_bytes))

		ret = -EFAULT;

	if (ret)

		goto out;

	cf_diag_ctrset.ctrset = start.counter_sets;

	ret = cf_diag_all_start();

out:

	if (ret)

		cf_diag_ctrset_clear();

	debug_sprintf_event(cf_diag_dbg, 2, "%s sets %#lx need %ld ret %d\n",

			    __func__, cf_diag_ctrset.ctrset, need, ret);

	return ret;

}



static long cf_diag_ioctl(struct file *file, unsigned int cmd,

			  unsigned long arg)

{

	int ret;



	debug_sprintf_event(cf_diag_dbg, 2, "%s cmd %#x arg %lx\n", __func__,

			    cmd, arg);

	get_online_cpus();

	mutex_lock(&cf_diag_ctrset_mutex);

	switch (cmd) {

	case S390_HWCTR_START:

		ret = cf_diag_ioctl_start(arg);

		break;

	case S390_HWCTR_STOP:

		ret = cf_diag_ioctl_stop();

		break;

	case S390_HWCTR_READ:

		ret = cf_diag_ioctl_read(arg);

		break;

	default:

		ret = -ENOTTY;

		break;

	}

	mutex_unlock(&cf_diag_ctrset_mutex);

	put_online_cpus();

	debug_sprintf_event(cf_diag_dbg, 2, "%s ret %d\n", __func__, ret);

	return ret;

}



static const struct file_operations cf_diag_fops = {

	.owner = THIS_MODULE,

	.open = cf_diag_open,

	.release = cf_diag_release,

	.unlocked_ioctl	= cf_diag_ioctl,

	.compat_ioctl = cf_diag_ioctl,

	.llseek = no_llseek

};



static struct miscdevice cf_diag_dev = {

	.name	= S390_HWCTR_DEVICE,

	.minor	= MISC_DYNAMIC_MINOR,

	.fops	= &cf_diag_fops,

};



static int cf_diag_online_cpu(unsigned int cpu)

{

	struct cf_diag_call_on_cpu_parm p;



	mutex_lock(&cf_diag_ctrset_mutex);

	if (!cf_diag_ctrset.ctrset)

		goto out;

	p.sets = cf_diag_ctrset.ctrset;

	cf_diag_ioctl_on(&p);

out:

	mutex_unlock(&cf_diag_ctrset_mutex);

	return 0;

}



static int cf_diag_offline_cpu(unsigned int cpu)

{

	struct cf_diag_call_on_cpu_parm p;



	mutex_lock(&cf_diag_ctrset_mutex);

	if (!cf_diag_ctrset.ctrset)

		goto out;

	p.sets = cf_diag_ctrset.ctrset;

	cf_diag_ioctl_off(&p);

out:

	mutex_unlock(&cf_diag_ctrset_mutex);

	return 0;

}



/* Initialize the counter set PMU to generate complete counter set data as

 * event raw data. This relies on the CPU Measurement Counter Facility device

 * already being loaded and initialized.
@@ -685,21 +1165,43 @@ static int __init cf_diag_init(void) 
		return -ENOMEM;

	}



	rc = misc_register(&cf_diag_dev);

	if (rc) {

		pr_err("Registration of /dev/" S390_HWCTR_DEVICE

		       "failed rc=%d\n", rc);

		goto out;

	}



	/* Setup s390dbf facility */

	cf_diag_dbg = debug_register(KMSG_COMPONENT, 2, 1, 128);

	if (!cf_diag_dbg) {

		pr_err("Registration of s390dbf(cpum_cf_diag) failed\n");

		return -ENOMEM;

		rc = -ENOMEM;

		goto out_dbf;

	}

	debug_register_view(cf_diag_dbg, &debug_sprintf_view);



	rc = perf_pmu_register(&cf_diag, "cpum_cf_diag", -1);

	if (rc) {

		debug_unregister_view(cf_diag_dbg, &debug_sprintf_view);

		debug_unregister(cf_diag_dbg);

		pr_err("Registration of PMU(cpum_cf_diag) failed with rc=%i\n",

		       rc);

		goto out_perf;

	}

	rc = cpuhp_setup_state_nocalls(CPUHP_AP_PERF_S390_CFD_ONLINE,

				       "perf/s390/cfd:online",

				       cf_diag_online_cpu, cf_diag_offline_cpu);

	if (!rc)

		goto out;



	pr_err("Registration of CPUHP_AP_PERF_S390_CFD_ONLINE failed rc=%i\n",

	       rc);

	perf_pmu_unregister(&cf_diag);

out_perf:

	debug_unregister_view(cf_diag_dbg, &debug_sprintf_view);

	debug_unregister(cf_diag_dbg);

out_dbf:

	misc_deregister(&cf_diag_dev);

out:

	return rc;

}

arch_initcall(cf_diag_init);
 
device_initcall(cf_diag_init);